Neotropy Lab · coming soon

Agents you can test, trace, and trust.

Build, test, and deploy custom agent behaviors with strict sandboxing — every tool call typed, every side effect declared, every session replayable.

Get Notified at Launch

01 — Playground

Poke the sandbox

Select a node to inspect its config, pick a scenario, and run the trace. Flip the sandbox off to see exactly what it is worth.

  • no PII export
  • spend cap $50 / action
  • network allowlist only

plannerintent=refund · confidence 0.94

toolrefunds_api.check(order_8841) → eligible

guardrailspend cap $50 · request $32 · PASS

respondrefund issued · cited policy §4.2

tracesaved · 4 spans · 412 ms

typed tool graph mcp servers as tools replayable eval traces

02 — Capabilities

From behavior sketch to deployed agent

Tool calling & MCP integration

Native function calling with MCP servers as first-class tools — schemas typed at compile time, versions pinned, side effects declared.

Multi-agent orchestration

Compose planners, critics, and workers on one typed graph with budgets, escalation paths, and human-in-the-loop gates.

Observability & evaluation traces

Token-level traces on every session, replayable offline evals, and graders that score production behavior against your rubric.

ephemeral filesystem
network allowlist
declared side effects
per-call audit log

coming soon

AI Lab — not live yet — but it is close.

We are finishing hardening, safety review, and onboarding tooling. Leave your email and we will notify you the moment it opens — one email, at launch, nothing else.

SOC 2 Type II on-prem available data never leaves your VPC